[ANNOUNCE] makedumpfile: Postpone the release of 1.5.1
kumagai-atsushi at mxc.nes.nec.co.jp
Thu Nov 29 03:32:50 EST 2012
On Tue, 27 Nov 2012 21:32:34 +0900 (JST)
HATAYAMA Daisuke <d.hatayama at gmail.com> wrote:
> From: Atsushi Kumagai <kumagai-atsushi at mxc.nes.nec.co.jp>
> Subject: [ANNOUNCE] makedumpfile: Postpone the release of 1.5.1
> Date: Tue, 27 Nov 2012 19:37:27 +0900
> > Hello,
> >> If there is no problem, I will release v1.5.1 GA on Nov 28.
> > I did regression test for v1.5.1 in the last weekend and found some bugs.
> > So the release date will be delayed by few weeks, sorry.
> > However, I don't think there are big issues for almost all users, because:
> > - #1 doesn't happen unless users specify -x option, and it's unnecessary
> > for these kernels.
> I don't think so. pageflags was rewriten as enumeration on v2.6.26. So
> debuginfo of the related kernels contains correct PG_buddy value. If
> passing vmlinux with -x option, makedumpfile should work well. I think
> problematic case happens only through vmcoreinfo with -i option.
Sorry, it was my misunderstanding. You are right.
> > - As for #2, I think there are few users.
> > So, I think v1.5.1-rc is enough for almost all users.
> > BUGS:
> > #1. Fail to exclude free pages with page_is_buddy_v2() in some cases.
> > I mentioned this issue in the post below:
> > http://lists.infradead.org/pipermail/kexec/2012-November/007404.html
> > Related kernel: v2.6.27 to v2.6.37
> This is a bit wrong. Correct one is: v2.6.30 to v2.6.37.
I wrote there versions as the version which has the possibility of this
issue happening. In other words, the version which uses CONFIG_PAGEFLAGS_EXTENDED.
So, "v2.6.27 to v2.6.37" was my mistake after all.
Anyway, correct one is: v2.6.26 to v2.6.37.
I'll avoid this issue by remove hard-corded values for PG_buddy.
> > #2. Fail to exclude free pages with page_is_buddy_v1() in some cases.
> > I'm investigating this issue now.
> > Related kernel: v2.6.15 to v2.6.17
> As I noted in its patch description, I have yet to test on these
> kernel versions. Thanks for your testing.
I think the release of v1.5.1 shouldn't be delayed with this issue because
as you said below.
So, I'll just revert "[PATCH v2 09/10] Add page_is_buddy for old kernels"
to avoid this issue. This means that we always use free_list logic for
v2.6.15 to v2.6.17.
> But I don't have enough reason to support these kernels now. These
> kernels themselves have bug on handling buddy pages, so PG_buddy was
> introduced during v2.6.18-rc to fix the bug. Even if there is
> distribution based on these kerenel versions, some kind of fixing
> patch introducing PG_buddy would be backported and that must be. In
> particular, I don't think machine with huge memory uses such buggy
> HATAYAMA, Daisuke
More information about the kexec